arbazkhan971/secure

Activate When

• /godmode:secure, "security audit", "vulnerabilities"
• "harden", "pentest", "threat model"
• "sql injection", "owasp", "owasp top 10", "vulnerability scan"
• "check owasp", "owasp issues", "security issues"

Workflow

1. Recon

# Dependency audit
npm audit 2>/dev/null || pip audit 2>/dev/null \
  || cargo audit 2>/dev/null

# Scan for hardcoded secrets
grep -rn 'SECRET\|API_KEY\|PASSWORD\|PRIVATE_KEY' \
  --include="*.ts" --include="*.py" --include="*.go" \
  --include="*.env" src/ 2>/dev/null | head -20

# List all public routes
grep -rn "router\.\(get\|post\|put\|delete\)" \
  --include="*.ts" --include="*.js" -l 2>/dev/null

2. Asset Map

List each: DB (type+version), auth mechanism (JWT/session/OAuth + expiry), external APIs, all <input>/<form> elements, public routes.

3. Trust Boundaries

Draw: client<->server, public<->auth, user<->admin, svc<->svc, CI<->prod, internal<->external. Each boundary = attack surface.

4. STRIDE Analysis

For each boundary: Spoofing, Tampering, Repudiation, Info Disclosure, DoS, Elevation.

5. Iterate

categories = OWASP_TOP_10 + STRIDE  # 16 total
current_iteration = 0
WHILE untested categories remain:
    current_iteration += 1
    Pick untested category.
    Priority: Injection > Broken Auth > XSS
      > SSRF > IDOR > remaining.
    Test as 4 personas:
      P1: External (no auth)
      P2: Insider (valid session)
      P3: Supply Chain (malicious dep)
      P4: Infrastructure (server access)
    Each finding: file:line + exploit steps
      + severity + fix (code snippet).
    Every 5 iters: print progress.

IF finding has code evidence from 1+ personas: KEEP
IF no code evidence from any persona: DISCARD

6. Red-Team Persona Voting

FOR each finding:
  Each persona rates: EXPLOITABLE (1) or NOT (0)
  Score = sum (0-4):
    4/4 = CRITICAL (unanimous)
    3/4 = HIGH (likely exploitable)
    2/4 = MEDIUM (conditional)
    1/4 = LOW (edge case)
    0/4 = DISCARD (theoretical)

7. Report

Print: OWASP {N}/10, STRIDE {N}/6. {C}C {H}H {M}M {L}L. PASS if 0 critical + 0 high. Else FAIL.

8. Auto-Fix (if --fix)

For Critical/High: fix -> commit -> run full test suite -> revert if ANY test breaks.

Findings Format

Every finding MUST use:

SEVERITY|FILE:LINE|DESCRIPTION|FIX

Examples:

CRITICAL|src/api/auth.ts:42|SQL injection via
  unsanitized email|Use parameterized query
HIGH|src/middleware/cors.ts:8|CORS wildcard|
  Set explicit allowed origins
MEDIUM|src/utils/crypto.ts:15|Math.random() for
  session token|Use crypto.randomBytes(32)

TSV Logging

Log to .godmode/security-findings.tsv: iteration\tcategory\tpersona\tseverity\tfile_line\tdescription\tfix\tstatus\tvotes

<!-- tier-3 -->

Quality Targets

• Critical CVEs in deps: <1 allowed
• SAST scan: <5min per 100KB source
• Hardcoded secrets: <1 detected per scan

Hard Rules

0. Inherits Default Activations per SKILL.md §14. Principles prelude, pre-commit audit, terse/stdio/tokens, DispatchContext validation, Progressive Disclosure routing, discard cost hierarchy all fire by default.
1. Every finding: file:line + exploit + proof.
2. Cover all OWASP Top 10 x 4 personas = 40 minimum.
3. Never approve with Critical findings open.
4. No code evidence from any persona = DISCARD.
5. Auto-fix must run full test suite after each fix.
6. Never ask to continue. Loop autonomously.

Keep/Discard Discipline

KEEP if: file:line evidence AND 1+ persona
  rates EXPLOITABLE AND steps reproducible
DISCARD if: no code evidence OR 0/4 votes
  OR duplicates existing finding
  Every discard logged with reason.

Overfitting Prevention

Every finding must be exploitable in the general case, not only in this specific commit state. If the finding depends on a transient condition → classify as noise.

Stop Conditions

STOP when FIRST of:
  - All OWASP + STRIDE categories tested
  - max_iterations reached
  - 3 consecutive boundaries produce 0 findings
    AND coverage >80%
  - >5 discards with no actionable replacements

Error Recovery

IF category produces no findings after analysis:
  log NO_FINDING with justification, move on
IF scanning tool fails:
  retry once, then fall back to manual review
IF finding cannot be reproduced:
  downgrade to INFO, tag UNVERIFIED
  Max 2 re-verification attempts

Output Format

Print: Secure: OWASP {N}/10, STRIDE {N}/6. {findings} findings. {kept} kept, {discarded} discarded. Status: {DONE|PARTIAL}.